By Sunghack Lim, Professor in the Department of International Relations at the University of Seoul. He holds a Ph.D. in Political Science from Pennsylvania State University. Currently, he serves as a Member of the AI and Democracy Subcommittee under the Presidential Council on National AI Strategy, a Member of the Indo-Pacific Democracy Forum, and an Advisory Committee Member for the Ministry of Foreign Affairs. 

On Invitation for LI’s Reimagined Newsletter – The Liberal Dispatch

Sign Up for The Liberal Dispatch

South Korea’s local elections in June 2026 showed both sides of democracy’s AI future in a single month. By late May, the National Election Commission had asked platforms to take down 10,319 deepfake videos tied to the campaign. That is about 25 times the number from the general election just two years before. Fake speeches, cloned voices, and manipulated images made up a third of all the election-law violations police caught.

And yet, in the same election, a young district council candidate ran his whole campaign with an “AI campaign manager” app. It mapped foot traffic, planned his canvassing routes, and even built his website. He spent about 3 million won — roughly USD 2,200 — and hired no staff at all. The technology that filled the information space with fakes also let a newcomer with no money and no machine politics behind him get into the race.

This is why liberals should be wary of the two stories usually told about AI. One camp sees it as a threat to democracy and wants to ban it. The other treats any rule as a brake on innovation and tells citizens to trust the market. Neither fits our tradition. Liberal International’s Andorra Manifesto puts it plainly: human creativity is close to unlimited, and our job is to stop clear abuses “through transparent supervision that does not unduly hinder the process of scientific discovery.” We do not have to pick between innovation and democracy. We have to build institutions where one supports the other.

That work starts by being clear about where AI actually does its damage. Three levels help. At the level of the individual citizen, the question is whether people can still judge for themselves what is true. Korea shows how fragile this is. Generative AI now reaches 78 percent of the population, and almost half of users say they trust what it tells them. But in a government survey, 42 percent of people who had seen a deepfake admitted they could not tell it from the real thing. Heavy reliance sitting on top of weak detection is exactly the ground where the “liar’s dividend” grows. Once anything might be fake, even real evidence of wrongdoing can be waved away as a fake.

At the level of social groups, the question is equality. Old digital gaps are turning into AI gaps. Older people, people with disabilities, and low-income citizens use AI at barely half

the national rate, even as governments lean more and more on data those same groups hardly show up in.

At the level of institutions, the question is whether elections stay honest and whether officials stay answerable when algorithms help them decide.

Judged against these three tests, even a digitally advanced democracy like Korea comes up short. Of the government’s 9.9 trillion won AI budget for 2026, about 98 percent goes to industry and national “AI transformation.” Roughly 2 percent goes to digital inclusion, AI literacy, and support for workers whose jobs are exposed. National AI ethics principles exist, which is good, but they only encourage — they cannot enforce. This lopsidedness is not only a Korean problem. It is what AI policy looks like almost everywhere: competitiveness first, citizens later.

A liberal AI policy would keep all that ambition and finish the job it leaves undone. It comes down to four things. First, transparency instead of censorship: label AI-made political content and fund public tools people can use to check what they see, so the state never becomes the sole judge of what is true. Second, fair data: require public-sector AI in welfare, hiring, and policing to be checked for who its data leaves out, so it does not quietly automate the inequalities we already have. Third, election rules that tell manipulation apart from participation: ban deceptive synthetic content, false voting information, and hidden micro-targeting, while protecting the ordinary uses that let a candidate with 3 million won stand against a party machine. Fourth, an independent watchdog with real teeth — one that can audit systems, ask how they work, and shut down the harmful ones — plus a citizen’s right to an explanation, an appeal, and a human second look. As AI turns “agentic” and starts carrying out tasks on its own rather than just producing text, those rights are what separate a helpful tool from an unaccountable ruler.

The stakes reach past any one country. Across Asia, digital authoritarianism — AI-driven surveillance and tight control of information used to keep power — is a model other governments can copy. Democracies that either ban the technology or use it without any accountability will not hold that model back. What holds it back is example: proof that a society can be advanced and free at the same time. Korea, with its strong digital infrastructure and lively civil society, is well placed to try — but no democracy pulls this off alone.

Liberals have never believed that progress and freedom are enemies. AI is our chance to

show it, as long as we remember that democracy’s health in this era will turn not on how fast the technology moves, but on how well we govern it.